Summary form only given. The authors describe a rule that can optimally allocate bits in the sense that (weighted) mean-squared errors are minimized for the set of k//i-level quantizers (i equals 0, 1, . . . , Q-1) with Q and k//i's prespecified. The optimality, undominatedness, and uniqueness of the optimum allocation are proved, and the theoretical results are supported by some examples. The algorithm has a simple quantizerlike structure, where inputs are the logarithms of component variances and decision levels are simply calculated using the normalized distortion rate function values of an encoder at b//i equals log//2 k//i bits, i equals 0, 1, . . . , Q-1.
|Original language||English (US)|
|Number of pages||1|
|State||Published - Dec 1 1986|
ASJC Scopus subject areas